[pve-devel] [PATCH qemu-server] Fix: cpu hotplug feature can't be changed online

[pve-devel] [PATCH qemu-server] Fix: cpu hotplug feature can't be changed online

[Alexandre Derumier]

The cpus are passed as devices with specific id only when cpu hotplug is enable
at start.
We can't enable/disable it online or vcpu hotplug api will thrown errors
not finding core id.

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
---
 PVE/QemuServer.pm | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm
index 2895675..22d1c71 100644
--- a/PVE/QemuServer.pm
+++ b/PVE/QemuServer.pm
@@ -4961,7 +4961,7 @@ sub vmconfig_hotplug_pending {
 	my $force = $pending_delete_hash->{$opt}->{force};
 	eval {
 	    if ($opt eq 'hotplug') {
-		die "skip\n" if ($conf->{hotplug} =~ /memory/);
+		die "skip\n" if ($conf->{hotplug} =~ /(cpu|memory)/);
 	    } elsif ($opt eq 'tablet') {
 		die "skip\n" if !$hotplug_features->{usb};
 		if ($defaults->{tablet}) {
@@ -5022,6 +5022,7 @@ sub vmconfig_hotplug_pending {
 	eval {
 	    if ($opt eq 'hotplug') {
 		die "skip\n" if ($value =~ /memory/) || ($value !~ /memory/ && $conf->{hotplug} =~ /memory/);
+		die "skip\n" if ($value =~ /cpu/) || ($value !~ /cpu/ && $conf->{hotplug} =~ /cpu/);
 	    } elsif ($opt eq 'tablet') {
 		die "skip\n" if !$hotplug_features->{usb};
 		if ($value == 1) {
-- 
2.39.2

Re: [pve-devel] [PATCH qemu-server] Fix: cpu hotplug feature can't be changed online

[Fiona Ebner]

Am 10.10.23 um 17:37 schrieb Alexandre Derumier:
> The cpus are passed as devices with specific id only when cpu hotplug is enable
> at start.
> We can't enable/disable it online or vcpu hotplug api will thrown errors
> not finding core id.

When removing cores after enabling the option this is true, but I can
1. start a VM without CPU hotplug and fewer than max vCPUs
2. enable CPU hotplug
3. add more cores

And doing the disable of the option online also doesn't seem problematic
at a first glance. So I thought this would technically be a breaking change.

But actually, the change is completely justified, because of migration.
Because the QEMU commandline changes based on the hotplug setting, so
the source and target VM will not agree and loading the state on the
target will get confused and crash:

> Oct 11 14:08:23 pve8a2 QEMU[160882]: kvm: get_pci_config_device: Bad config data: i=0x9a read: 8 device: 3 cmask: ff wmask: 0 w1cmask:0
> Oct 11 14:08:23 pve8a2 QEMU[160882]: kvm: Failed to load PCIDevice:config
> Oct 11 14:08:23 pve8a2 QEMU[160882]: kvm: Failed to load virtio-scsi:virtio
> Oct 11 14:08:23 pve8a2 QEMU[160882]: kvm: error while loading state for instance 0x0 of device '0000:00:05.0/virtio-scsi'
> Oct 11 14:08:23 pve8a2 QEMU[160882]: kvm: load of migration failed: Invalid argument

Therefore,
Reviewed-by: Fiona Ebner <f.ebner@proxmox.com>

Re: [pve-devel] [PATCH qemu-server] Fix: cpu hotplug feature can't be changed online

[DERUMIER, Alexandre]

Am 10.10.23 um 17:37 schrieb Alexandre Derumier:
> The cpus are passed as devices with specific id only when cpu hotplug
> is enable
> at start.
> We can't enable/disable it online or vcpu hotplug api will thrown
> errors
> not finding core id.

>>When removing cores after enabling the option this is true, but I can
>>1. start a VM without CPU hotplug and fewer than max vCPUs
>>2. enable CPU hotplug
>>3. add more cores

>>And doing the disable of the option online also doesn't seem
>>problematic
>>at a first glance. So I thought this would technically be a breaking
>>change.

Note that is also breaking unplug of cores present before enable
hotplug

>>1. start a VM without CPU hotplug and fewer than max vCPUs
>>2. enable CPU hotplug
>>3. remove cores



(But yes, the more important is migration breaking)

[pve-devel] applied: [PATCH qemu-server] Fix: cpu hotplug feature can't be changed online

[Thomas Lamprecht]

Am 10/10/2023 um 17:37 schrieb Alexandre Derumier:
> The cpus are passed as devices with specific id only when cpu hotplug is enable
> at start.
> We can't enable/disable it online or vcpu hotplug api will thrown errors
> not finding core id.
> 
> Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
> ---
>  PVE/QemuServer.pm | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
>

applied, thanks!